fix(ci): 🐛 let the C++ MPI test variants oversubscribe under OpenMPI 5

[diagonal-hamiltonian]

🤖 AI text below 🤖

Summary

The C++ mpi ctest variants spawn their own mpiexec -n <n> from inside the test, so the --map-by :OVERSUBSCRIBE that the neighbouring Python steps pass on the command line never reaches them — only the environment can. The variable both justfile recipes already export, OMPI_MCA_rmaps_base_oversubscribe, is the OpenMPI 4 spelling and OpenMPI 5 ignores it.

Measured against OpenMPI 5.0.8, mpiexec -n 4 --host localhost:2 hostname:

env	result
none	rc=1 — refuses to place
OMPI_MCA_rmaps_base_oversubscribe=1	rc=1 — still refuses
PRTE_MCA_rmaps_default_mapping_policy=:oversubscribe	rc=0, 4 ranks

This is latent on main, where monoprop_MPI_TEST_PROCS is "2" and every runner can place 2 ranks — CI stays green either way. It bites any branch that widens the sweep: at 4 ranks the ubuntu-26.04 and macos-15 lanes fail with prte-rmaps-base:alloc-error / "not enough slots", while ubuntu-26.04-arm places it. Both test.yml and the Collect coverage [mpi:on] job are affected, since the coverage recipe runs the same ctest --label-regex mpi.

Found while rebasing #296, which widens the sweep to "2;4" deliberately — at R=2 the peer plan resolves only one peer, so every for k in [0, f) in the sparse transport stays a single iteration and the multi-peer paths never run.

Changes

• .github/workflows/test.yml: set PRTE_MCA_rmaps_default_mapping_policy (and keep the OpenMPI 4 name) on the Run C++ unit tests under MPI step, which previously set no environment at all.
• justfile: add the same export to test-mpi and to code-coverage-collect, both of which run ctest --label-regex mpi after exporting only the OpenMPI 4 name.

Checklist

• Tests added or updated to cover the changes — n/a; this is CI configuration. Verified by direct measurement against OpenMPI 5.0.8 (table above). Note it cannot be proven by this PR's own CI, because main runs the sweep at 2 ranks where the bug is invisible; perf(mpi): 🧭 GF(2)-linear rank routing — messages per rank flat in R #296 exercises it at 4.
• Documentation updated (docstrings, docs/, CONTRIBUTING.md) if needed — n/a
• CHANGELOG / release notes updated if applicable — n/a

AI/LLM disclosure

• I used the following tool to help write this PR description: Claude Code (claude-opus-5)
• I used the following tool to generate or modify code: Claude Code (claude-opus-5)

[github-actions]

Docs preview: https://pr-324.monoprop-docs.pages.dev

[robertodr]

thanks good catch! Does OpenMPI 4 honor the PRTE_MCA_* version?

[diagonal-hamiltonian]

thanks good catch! Does OpenMPI 4 honor the PRTE_MCA_* version?

Don't know sorry

[robertodr]

thanks good catch! Does OpenMPI 4 honor the PRTE_MCA_* version?

Don't know sorry

I checked, it does not. There was a breaking change in runtime between the versions

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.70%. Comparing base (a24143f) to head (a2ccda4).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #324   +/-   ##
=======================================
  Coverage   97.70%   97.70%           
=======================================
  Files          14       14           
  Lines         742      742           
  Branches       98       98           
=======================================
  Hits          725      725           
  Misses         12       12           
  Partials        5        5           

Flag	Coverage Δ
cpp	97.70% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud